Paper: VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation link
I will fix bugs in the future, here are my TODO list.
- implement and debug the network
- use naive loss to overfit the random dataset
- implement the evaluation(ADE, DE, ..)
- implement the loss
- implement the dataset interface
- running on the whole dataset
- submit result to the website https://eval.ai/web/challenges/challenge-page/454/overview
- visualize the result
- node completion
- docker environment
OS: Linux or MacOS. Because Argoverse dataset only can run on the Linux or MacOS.
See main.py.
TODO